That's not necessarily a good idea. Small vector uses rely on having nearly no overhead to be faster, so they can't use a generic system.
Of course if you need less than a VLEN-sized vector you're wasting throughput, but that applies just as much when using 128-bit vectors on AVX-capable hardware, and even worse so on AVX-512-capable (which, while double-pumped or equivalent to some extent on most impls, still has 512-bit-exclusive throughput on most).